Ambulance avoidance method and device, vehicle and storage medium

Abstract

The invention relates to the technical field of vehicles, in particular to an ambulance avoidance method and device, a vehicle and a storage medium, and the method comprises the steps: obtaining image data and sound data of a current vehicle, obtaining a first confidence coefficient and a second confidence coefficient according to the image data and the sound data, carrying out the fusion of the first confidence coefficient and the second confidence coefficient, and obtaining a fusion confidence coefficient; under the condition that the fusion confidence coefficient is greater than a preset fusion threshold value, obtaining current vehicle speed and current road condition information, and when it is determined that the current vehicle meets a preset lane changing avoidance condition according to the current vehicle speed and the current road condition information, inputting the current road condition information into a preset path planning model to obtain a target planning path, and controlling the current vehicle to avoid according to the target planned path. Therefore, the problems of low ambulance recognition reliability and inflexible avoidance strategy caused by single-mode sensing are solved, multi-mode sensing is fused, the dynamic path planning capability is achieved, the response is efficient, and the safe avoidance capability in a complex traffic environment can be improved.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a method, device, vehicle, and storage medium for ambulance obstacle avoidance. Background Technology

[0002] With the rapid development of autonomous driving technology, intelligent vehicles have demonstrated enormous potential in improving traffic efficiency and driving safety. In complex urban traffic environments, autonomous vehicles need to possess the ability to quickly identify and appropriately avoid emergency rescue vehicles to ensure public safety and comply with traffic regulations. Ambulances, as the core force in emergency rescue, typically activate their sirens and flashing lights when on duty, requesting surrounding vehicles to give way.

[0003] Related technologies either utilize image recognition algorithms to achieve visual perception or employ microphone arrays to collect sound signals and perform acoustic recognition by analyzing the spectral characteristics of the siren. In terms of avoidance strategies, related technologies often employ preset rules, such as immediately slowing down and stopping on the right after detecting an ambulance, or performing a simple lane change operation when it is safe to do so in the adjacent lane.

[0004] However, the above technologies still have significant drawbacks: First, recognition methods that rely solely on visual or auditory modalities are not robust and are prone to false detections or missed detections at night, in rain or fog, in tunnels, or in environments with strong noise; second, the avoidance strategies in the relevant technologies lack flexibility and cannot be dynamically adjusted according to actual road conditions, leading to avoidance failures or safety hazards. Therefore, there is an urgent need for an autonomous driving avoidance system that integrates multimodal perception, has dynamic path planning capabilities, and is highly responsive, in order to improve the ability to safely yield in complex traffic environments. Summary of the Invention

[0005] This application provides a method, device, vehicle, and storage medium for ambulance avoidance, in order to solve the problems of low reliability of ambulance identification and inflexible avoidance strategies caused by single-modal perception. This application integrates multimodal perception, has dynamic path planning capabilities, and is highly responsive, which can improve the ability to safely yield in complex traffic environments.

[0006] The first aspect of this application provides a method for an ambulance to avoid a collision, comprising the following steps: Acquire the current vehicle's image and audio data; The image data is input into a preset ambulance image recognition model to obtain a first confidence level, and the sound data is input into a preset ambulance sound recognition model to obtain a second confidence level. The first confidence level and the second confidence level are fused to obtain a fused confidence level, and it is determined whether the fused confidence level is greater than a preset fusion threshold. If the fusion confidence is greater than the preset fusion threshold, the current vehicle speed and current road condition information are obtained. Based on the current vehicle speed and current road condition information, it is determined whether the current vehicle meets the preset lane change and avoidance conditions. When the current vehicle meets the preset lane change and avoidance conditions, the current road condition information is input into the preset path planning model to obtain the target planning path, and the current vehicle is controlled to avoid the obstacle using the target planning path.

[0007] Optionally, in some embodiments, determining whether the current vehicle meets the preset lane-changing and yielding conditions based on the current vehicle speed and the current road condition information includes: The minimum lane change distance is calculated based on the current vehicle speed. Based on the current road condition information, it is determined whether the lane line between the current vehicle and the adjacent lane is a dashed line, and it is determined whether the distance between the current vehicle and the target obstacle is greater than the minimum safe distance. When the lane change distance is greater than the minimum lane change distance, the lane line between the current vehicle and the adjacent lane is the dashed line, and the distance between the current vehicle and the target obstacle is greater than the minimum safe distance, the current vehicle is determined to meet the preset lane change and avoidance conditions.

[0008] Optionally, in some embodiments, determining whether the fusion confidence level is greater than a preset fusion threshold further includes: If the fusion confidence is less than or equal to the preset fusion threshold, millimeter-wave radar-assisted verification is enabled. If the millimeter-wave radar-assisted verification passes, the acquisition frame rate of the image data is increased, and / or the audio analysis window of the sound data is extended to obtain new image data and new sound data. A new first confidence level and a new second confidence level are obtained based on the new image data and the new sound data, and a new fusion confidence level is obtained based on the new first confidence level and the new second confidence level, and it is determined whether the new fusion confidence level is greater than the preset fusion threshold.

[0009] Optionally, in some embodiments, before inputting the current traffic information into the preset path planning model to obtain the target planned path, the following steps are included: Obtain historical path data and construct a dataset based on the historical path data; Based on a preset partitioning ratio, the dataset is divided into a training set, a validation set, and a test set. Construct a target neural network by inputting the training set into the target neural network for training to obtain initial model parameters; Based on the initial model parameters, the validation set is input into the target neural network for performance evaluation, and the initial model parameters are adjusted according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters. Based on the optimal model parameters, the test set is input into the target neural network for model testing, and when the test results meet the preset requirements, the preset path planning model is obtained.

[0010] A second aspect of this application provides a collision avoidance device for an ambulance, comprising: The acquisition module is used to acquire the image and sound data of the current vehicle; The input module is used to input the image data into a preset ambulance image recognition model to obtain a first confidence level, and input the sound data into a preset ambulance sound recognition model to obtain a second confidence level, fuse the first confidence level and the second confidence level to obtain a fused confidence level, and determine whether the fused confidence level is greater than a preset fusion threshold. The obstacle avoidance module is used to obtain current vehicle speed and current road condition information when the fusion confidence is greater than the preset fusion threshold, determine whether the current vehicle meets the preset lane change and obstacle avoidance conditions based on the current vehicle speed and the current road condition information, and when the current vehicle meets the preset lane change and obstacle avoidance conditions, input the current road condition information into the preset path planning model to obtain the target planned path, and control the current vehicle to avoid obstacles using the target planned path.

[0011] Optionally, in some embodiments, the avoidance module includes: The judgment unit is used to calculate the minimum lane change distance based on the current vehicle speed, determine whether the lane line between the current vehicle and the adjacent lane is a dashed line based on the current road condition information, and determine whether the distance between the current vehicle and the target obstacle is greater than the minimum safe distance. The determination unit is used to determine that the current vehicle meets the preset lane change and avoidance conditions when the lane change distance is greater than the minimum lane change distance, the lane line between the current vehicle and the adjacent lane is the dashed line, and the distance between the current vehicle and the target obstacle is greater than the minimum safe distance.

[0012] Optionally, in some embodiments, the input module further includes: The data reacquisition unit is used to enable millimeter-wave radar-assisted verification when the fusion confidence is less than or equal to the preset fusion threshold. If the millimeter-wave radar-assisted verification passes, the acquisition frame rate of the image data is increased and / or the audio analysis window of the sound data is extended to obtain new image data and new sound data. The fusion unit obtains a new first confidence level and a new second confidence level based on the new image data and the new sound data, and obtains a new fusion confidence level based on the new first confidence level and the new second confidence level, and determines whether the new fusion confidence level is greater than the preset fusion threshold.

[0013] Optionally, in some embodiments, before inputting the current road condition information into the preset path planning model to obtain the target planned path, the avoidance module includes: An acquisition unit is used to acquire historical path data and construct a dataset based on the historical path data; A partitioning unit is used to divide the dataset into a training set, a validation set, and a test set based on a preset partitioning ratio. A construction unit is used to construct a target neural network, and to obtain initial model parameters by inputting the training set into the target neural network for training. The input unit is used to input the validation set into the target neural network for performance evaluation based on the initial model parameters, and adjust the initial model parameters according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters. The generation unit is used to input the test set into the target neural network based on the optimal model parameters to test the model, and when the test results meet the preset requirements, obtain the preset path planning model.

[0014] A third aspect of this application provides a vehicle, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the ambulance avoidance method as described in the above embodiments.

[0015] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the ambulance avoidance method as described in the above embodiments.

[0016] Therefore, by acquiring the current vehicle's image and sound data, the image data is input into a preset ambulance image recognition model to obtain a first confidence level, and the sound data is input into a preset ambulance sound recognition model to obtain a second confidence level. The first and second confidence levels are fused to obtain a fused confidence level. It is then determined whether the fused confidence level is greater than a preset fusion threshold. If the fused confidence level is greater than the preset fusion threshold, the current vehicle speed and current road condition information are acquired. Based on the current vehicle speed and current road condition information, it is determined whether the current vehicle meets the preset lane change and avoidance conditions. When the current vehicle meets the preset lane change and avoidance conditions, the current road condition information is input into a preset path planning model to obtain a target planned path, and the current vehicle is controlled to avoid the obstacle using the target planned path. This solves the problem of low reliability in ambulance recognition and inflexible avoidance strategies caused by single-modal perception. This application integrates multi-modal perception, has dynamic path planning capabilities, and has a high response efficiency, which can improve the safe yielding capability in complex traffic environments.

[0017] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0018] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of an ambulance avoidance method provided according to an embodiment of this application; Figure 2 This is a schematic diagram illustrating the principle of an improved YOLOv7 model according to an embodiment of this application; Figure 3 This is a flowchart illustrating the DS evidence theory fusion process according to an embodiment of this application; Figure 4 This is a schematic diagram illustrating the principle of dynamic obstacle avoidance according to an embodiment of this application; Figure 5 This is a schematic diagram of path replanning provided according to one embodiment of this application; Figure 6 This is a schematic diagram of a vehicle architecture employing an ambulance avoidance method according to an embodiment of this application; Figure 7 This is a block diagram of an ambulance avoidance device provided according to an embodiment of this application; Figure 8 This is a structural schematic diagram of a vehicle provided according to an embodiment of this application. Detailed Implementation

[0019] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0020] The following description, with reference to the accompanying drawings, describes an ambulance avoidance method, apparatus, vehicle, and storage medium according to embodiments of this application. Addressing the problems mentioned in the background art, such as low reliability of ambulance recognition and inflexible avoidance strategies due to single-modal perception, this application provides an ambulance avoidance method. In this method, image data and sound data of the current vehicle are acquired. The image data is input into a preset ambulance image recognition model to obtain a first confidence level, and the sound data is input into a preset ambulance sound recognition model to obtain a second confidence level. The first and second confidence levels are fused to obtain a fused confidence level. It is determined whether the fused confidence level is greater than a preset fusion threshold. If the fused confidence level is greater than the preset fusion threshold, the current vehicle speed and current road condition information are acquired. Based on the current vehicle speed and current road condition information, it is determined whether the current vehicle meets preset lane-change avoidance conditions. When the current vehicle meets the preset lane-change avoidance conditions, the current road condition information is input into a preset path planning model to obtain a target planned path, and the current vehicle is controlled to avoid the obstacle along the target planned path. This solves the problems of low reliability in ambulance recognition and inflexible avoidance strategies caused by single-modal perception. This application integrates multimodal perception, has dynamic path planning capabilities, and is highly efficient in response, which can improve the ability to safely yield in complex traffic environments.

[0021] Specifically, Figure 1 This is a flowchart illustrating a method for an ambulance to avoid obstacles, as provided in an embodiment of this application.

[0022] like Figure 1 As shown, the ambulance's avoidance method includes the following steps: In step S101, the image data and sound data of the current vehicle are acquired.

[0023] Specifically, embodiments of the present invention can deploy a multi-view camera system in a vehicle, including a rear main camera and side auxiliary cameras, covering a 360° field of view around the vehicle. The rear camera is responsible for long-distance target capture (up to 200 meters), and the side cameras use wide-angle lenses to fill in blind spot information. Before inputting image data into a preset ambulance image recognition model, image data processing can also be performed, including: dynamic range expansion (HDR), demosaicing, noise suppression (based on the BM3D algorithm), and geometric distortion correction, to ensure that the image quality input to the detection model meets the high-precision requirements. Furthermore, adaptive illumination compensation technology is introduced, which dynamically adjusts exposure parameters and local contrast enhancement strategies by analyzing the ambient light intensity in real time (such as tunnel entrances and nighttime scenes).

[0024] For sound data acquisition, a multi-microphone array can be used to achieve sound source localization and beamforming, suppressing environmental noise interference (such as traffic background noise and wind and rain noise). Mel-frequency spectral features are extracted through time-frequency analysis (STFT), and dynamic time warping (DTW) is introduced to match the pulse pattern of the ambulance siren. For example, a four-channel microphone array layout is used, with microphones arranged in a cross shape on the vehicle roof, with an optimized spacing of 15cm, supporting audio acquisition at a high sampling rate of 48kHz and a signal-to-noise ratio of ≥70dB. The array has omnidirectional pickup capability, covering the 360° acoustic space around the vehicle, ensuring that ambulance sirens approaching from any direction can be effectively captured.

[0025] In step S102, image data is input into a preset ambulance image recognition model to obtain a first confidence level, and sound data is input into a preset ambulance sound recognition model to obtain a second confidence level. The first confidence level and the second confidence level are fused to obtain a fused confidence level, and it is determined whether the fused confidence level is greater than a preset fusion threshold.

[0026] The preset fusion threshold can be set by the user, obtained through a limited number of experiments, or obtained through a limited number of computer simulations; no specific limitation is made here.

[0027] Specifically, such as Figure 2 As shown, this embodiment of the application can add a dedicated ambulance detection branch on the YOLOv7 architecture to independently predict key visual features of ambulances (such as rotating warning lights, reflective strips on the vehicle body, and medical markings). This branch extracts multi-scale features through cascaded convolutional layers and uses an attention mechanism (SE module) to enhance feature representation capabilities. To address the challenge of small target detection, a Feature Pyramid Network (FPN++) and an adaptive anchor box generation algorithm are introduced to dynamically optimize the detection accuracy of targets at different scales. During the training phase, mixed precision training (AMP) and distributed data parallelism (DDP) are employed, and synthetic datasets (including extreme scenes such as rain, fog, snow, and backlighting) are used to improve the model's generalization ability.

[0028] In addition, a hybrid model based on deep residual network (ResNet) and bidirectional LSTM is designed. The former is used for spectral spatial feature extraction, and the latter captures temporal dependencies. The model output is the siren type (ambulance, fire truck, police car) and its confidence. The training data covers a variety of siren modes (continuous siren, intermittent siren) and complex acoustic environments (city streets, highway tunnels).

[0029] like Figure 3 As shown, this embodiment of the application uses Dempster-Shafer (DS) evidence theory to fuse image and sound recognition results, defining the basic probability allocation function (BPA) as follows: m (ambulance) = C image .C audio ; m(non-ambulance) = (1-C image )·(1-C audio ); m (uncertain) = 1 - m (ambulance) - m (non-ambulance); Where Cimage is the first confidence level, i.e., the confidence level for image recognition, C audio Here, m(ambulance) represents the confidence level of sound recognition, m(non-ambulance) represents the joint evidence strength that both the image and sound support the target being an ambulance, m(non-ambulance) represents the joint evidence strength that neither the image nor the sound supports the target being an ambulance, and m(uncertain) represents the remaining uncertainty.

[0030] Preferably, in this embodiment of the invention, the threshold can be 0.8. When m (ambulance) ≥ 0.8, the system determines that an ambulance exists and triggers an avoidance decision. When m (non-ambulance) ≥ 0.8, it determines that there is no emergency vehicle. When both are below the threshold, or m (uncertain) is high, the system enters "observation mode" or relies on other sensors for further judgment.

[0031] Optionally, in some embodiments, determining whether the fusion confidence is greater than a preset fusion threshold further includes: if the fusion confidence is less than or equal to the preset fusion threshold, then enabling millimeter-wave radar-assisted verification; if the millimeter-wave radar-assisted verification passes, then increasing the acquisition frame rate of the image data and / or extending the audio analysis window of the sound data to obtain new image data and new sound data; obtaining a new first confidence and a new second confidence based on the new image data and the new sound data, and obtaining a new fusion confidence based on the new first confidence and the new second confidence, and determining whether the new fusion confidence is greater than the preset fusion threshold.

[0032] Specifically, in the multimodal fusion recognition process, when the initial fusion confidence of the image and sound is less than a preset threshold (e.g., 0.8), or m (non-ambulance) is greater than or equal to the preset threshold, or m (uncertain) is greater than the preset threshold, the system enters an "uncertain" state. To improve the reliability of the judgment, this solution introduces millimeter-wave radar as an auxiliary verification method. If the radar detects a rapidly approaching vehicle from the rear or side (e.g., relative speed greater than 15 km / h and distance less than 150 meters), it considers the possibility of an emergency vehicle to be high. The system will actively enhance its perception capabilities: increase the camera image acquisition frame rate (e.g., from 30fps to 60fps) and extend the audio analysis window (e.g., from 200ms to 500ms) to obtain higher resolution visual and auditory information. Based on the updated data, the image and sound confidence is recalculated, and DS fusion is performed again to form a new fusion confidence for secondary judgment. This mechanism realizes a closed-loop optimization of "perception-suspicion-enhancement-rejuvenation," significantly improving the robustness of recognition in complex scenarios.

[0033] This solves the problem of missed detection in harsh environments by single or dual-modal sensing. Through the "radar screening + sensing enhancement" mechanism, the sensing quality at critical moments is dynamically improved without increasing the continuous high computing power burden.

[0034] In addition, to improve the functional safety and fault tolerance of autonomous driving systems in the event of sensor failure, this application also provides a redundant perception switching mechanism. When the system detects a hardware failure in the camera or microphone (such as signal interruption, data anomaly, communication timeout, etc.), it automatically triggers a degraded operation mode and switches to a recognition scheme dominated by millimeter-wave radar. In this mode, the system uses radar point cloud data to cluster targets, combines multi-frame tracking algorithms to extract motion trajectory features (such as continuous high-speed driving, frequent lane changes, rapid approach to the vehicle, etc.), and identifies potential ambulances through a preset emergency vehicle behavior model. Although the lack of visual and acoustic features reduces the recognition accuracy, it can still maintain basic avoidance functions to ensure that the system does not fail. Thus, the redundancy mechanism significantly improves the reliability and safety of the system, maintaining basic recognition capabilities even when the main sensor fails, and preventing the complete loss of the obstacle avoidance function due to a single point of failure.

[0035] In step S103, if the fusion confidence is greater than the preset fusion threshold, the current vehicle speed and current road condition information are obtained. Based on the current vehicle speed and current road condition information, it is determined whether the current vehicle meets the preset lane change and avoidance conditions. When the current vehicle meets the preset lane change and avoidance conditions, the current road condition information is input into the preset path planning model to obtain the target planning path, and the current vehicle is controlled to avoid the obstacle using the target planning path.

[0036] Furthermore, in some embodiments, determining whether the current vehicle meets the preset lane change and avoidance conditions based on the current vehicle speed and current road condition information includes: calculating the minimum lane change distance based on the current vehicle speed; determining whether the lane line between the current vehicle and the adjacent lane is a dashed line based on the current road condition information; and determining whether the distance between the current vehicle and the target obstacle is greater than the minimum safe distance; when the lane change distance is greater than the minimum lane change distance, the lane line between the current vehicle and the adjacent lane is a dashed line, and the distance between the current vehicle and the target obstacle is greater than the minimum safe distance, it is determined that the current vehicle meets the preset lane change and avoidance conditions.

[0037] Specifically, in combination Figure 4 and Figure 5 As shown, after completing multimodal fusion recognition and confirming the presence of an ambulance (fusion confidence ≥ preset threshold), the system enters the dynamic avoidance decision-making stage. At this point, the system first acquires the vehicle's current driving status (such as speed and acceleration) and surrounding road condition information (including lane line type, adjacent lane obstacle distribution, traffic flow speed, high-precision map traffic restriction information, etc.) to determine whether safe lane-changing conditions are met. Lane-changing avoidance conditions typically include: adjacent lanes are dashed lines, sufficient safety distance (based on the "2-second rule" plus a safety margin), no rapidly approaching vehicles behind, and not being in a prohibited lane-changing area (such as a solid line area or construction section). If all conditions are met, the system inputs the current road condition information into a preset path planning model (such as an improved trajectory optimization model in the A* or Frenet coordinate system), generates a smooth, compliant, and safe target planning path, and performs lateral and longitudinal control through the drive-by-wire chassis to guide the vehicle along this path to complete the avoidance maneuver.

[0038] In actual implementation, the feasibility of lane changing can be assessed from the following aspects: (1) Safety distance model: based on vehicle speed (v) ego ) and reaction time (default 2 seconds), dynamically calculate the minimum safe distance between adjacent lanes: D safe =V ego *2+5m, where D safe For the minimum lane change distance, V ego This represents the current vehicle speed.

[0039] (2) Lane status analysis: Combine high-precision map to obtain lane line type (dashed / solid line), lane width and curvature information. If the adjacent lane is dashed and the obstacle distance is >D safe Then a lane change command will be generated.

[0040] In addition, the embodiments of this application can assess the feasibility of lane changing through a real-time scoring system. By constructing a quantitative scoring model in the range of 0 to 1, the system comprehensively evaluates three dimensions: lane compliance, obstacle risk, and traffic flow. The weighted calculation yields a lane changing feasibility score. When the score is ≥0.6, the system determines that the lane changing conditions are met and performs an avoidance action. This solution achieves dynamic, interpretable, and highly safe avoidance decisions, effectively balancing yielding needs and driving safety.

[0041] Optionally, in some embodiments, before inputting the current traffic information into a preset path planning model to obtain the target planned path, the method includes: acquiring historical path data and constructing a dataset based on the historical path data; dividing the dataset into a training set, a validation set, and a test set based on a preset partitioning ratio; constructing a target neural network, inputting the training set into the target neural network for training to obtain initial model parameters; based on the initial model parameters, inputting the validation set into the target neural network for performance evaluation, and adjusting the initial model parameters according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters; based on the optimal model parameters, inputting the test set into the target neural network for model testing, and obtaining the preset path planning model when the test results meet preset requirements.

[0042] Specifically, this application introduces a model training and validation process based on historical data before deployment. First, a large amount of historical path data from real-world driving scenarios (including vehicle status, surrounding traffic environment, driver lane-changing behavior, etc.) is collected to construct a structured dataset, which is then proportionally divided into training, validation, and test sets. Next, a target neural network (such as a trajectory prediction network based on Transformer or CNN-LSTM) is constructed, and initial training is performed using the training set to obtain initial model parameters. The model performance (such as trajectory prediction error and comfort indicators) is evaluated using the validation set, and hyperparameters and network weights are iteratively optimized until the joint loss function (such as position error + rate of curvature change) on the validation set converges, obtaining the optimal model parameters. Finally, the test set is used for final performance validation. When the avoidance success rate, path smoothness, and other indicators meet preset requirements, the model is solidified as a "preset path planning model" for real-world vehicle avoidance decisions, ensuring high reliability and driving comfort under complex road conditions.

[0043] It should be noted that the embodiments of this application also improve traditional algorithms in multiple dimensions to adapt to the path planning needs of autonomous vehicles in emergency avoidance scenarios. First, traffic rule weights are introduced into the cost function, setting high penalty values ​​for prohibited lane-changing areas, solid lines, construction sections, etc., to ensure that the generated path complies with traffic regulations; at the same time, a dynamic obstacle cost term is added, which calculates the risk cost in real time based on the speed, acceleration, and time to collision (TTC) of surrounding vehicles, improving path safety. Second, a multi-objective optimization strategy is adopted, using travel time, energy consumption (such as motor power consumption), and path deviation (the deviation relative to the desired lane) as optimization objectives. The non-dominated solution set is screened through Pareto front analysis, and the final decision is made in combination with driving style preferences or emergency level. Finally, a real-time update mechanism is established. The system refreshes the environmental perception data every 100ms. Once a sudden change in the ambulance trajectory (such as sudden acceleration or lane change) or the appearance of a new obstacle (such as a vehicle temporarily occupying the lane) is detected, the current planning is immediately interrupted and a replanning process is triggered to ensure that the avoidance path is always synchronized with the dynamic environment, achieving a safe, compliant, and efficient real-time response.

[0044] Furthermore, embodiments of this application can also design a state space (vehicle state, environmental information) and an action space (steering, icing and deceleration), with the reward function focusing on maximizing the safe distance and minimizing the path deviation, and the decision logic optimized through a policy gradient algorithm.

[0045] When controlling the current vehicle to avoid obstacles along the planned path, this application integrates a high-response steer-by-wire system with a cooperative control algorithm. At the hardware level, the steer-by-wire system possesses high-precision angle control capabilities, with a steering actuator response time of less than 20 milliseconds, ensuring the vehicle can quickly respond to path planning commands. The brake-by-wire system supports rapid pressure build-up and can trigger full braking in emergency situations, achieving a braking force greater than or equal to... The deceleration ensures longitudinal safety during the avoidance process. At the software level, the control algorithm adopts a hierarchical collaborative architecture: longitudinal control uses a PID speed regulation algorithm to adjust the motor or engine output in real time, accurately tracking the target vehicle speed, with speed error controlled within positive and negative limits. Within this range, improper acceleration and deceleration can affect the timing of avoidance maneuvers. Lateral control employs a pre-aiming tracking algorithm, dynamically predicting the forward viewpoint based on the curvature information of the target planned path and calculating the optimal steering angle command. While ensuring path tracking accuracy, it actively limits lateral acceleration to no more than 0.3g, improving ride comfort and roll safety. This achieves a high-fidelity reproduction from "planned path" to "vehicle execution," enabling the autonomous vehicle to quickly and smoothly complete lane changes and sidewalk maneuvers after recognizing an ambulance, and to brake rapidly in emergency situations, comprehensively improving the reliability of emergency response and the driving experience.

[0046] In actual implementation, such as Figure 6 As shown, this application can implement the ambulance avoidance method through the following vehicle module.

[0047] (1) Multimodal sensing module: Image recognition unit: Hardware: Rear-facing driving camera + side-facing fisheye camera. Algorithm: Improved YOLOv7 model, added ambulance feature detection head (police lights, crosshairs, reflective strips), supports nighttime infrared enhancement mode.

[0048] Sound Recognition Unit: Hardware: 4-microphone array (sampling rate 48kHz, signal-to-noise ratio ≥70dB). Algorithm: Combining Mel-spectrum mapping and LSTM network to distinguish ambulance sirens (frequency 450-1500Hz, pulse mode 0.5-2Hz).

[0049] Data fusion: The DS evidence theory is used to weight the confidence of the image and sound recognition results. When the threshold is ≥0.8, it is determined to be an ambulance.

[0050] 2. Dynamic Decision Module: Avoidance strategy generation: Lane change and avoidance: Based on map data built by autonomous driving, calculate the safe distance between adjacent lanes (formula: safe distance = vehicle speed × 2s + 5m).

[0051] Route replanning: If changing lanes is not feasible (e.g., lane lines are solid lines or there are obstacles in adjacent lanes), the A* algorithm is invoked to generate an alternative route, and the navigation system is updated synchronously.

[0052] Priority control: The avoidance action is optimized through reinforcement learning (PPO algorithm), with the objective function being to minimize path deviation and maximize safe distance.

[0053] 3. Execution Control Module: Hardware: Drive-by-wire chassis (steering accuracy 0.1°, braking response time <50ms).

[0054] Control logic: The path planning results are sent to the actuator via the CAN bus. The longitudinal control uses PID speed regulation, and the lateral control uses a preview tracking algorithm.

[0055] This application embodiment can construct multi-scenario test cases (such as urban roads, highways, and severe weather) in a virtual simulation platform. The evaluation metrics are: recognition accuracy, avoidance response time, and path tracking error. Additionally, complex traffic flow is simulated in a closed test field to verify the system's reliability in a real-vehicle environment. Key metrics: average avoidance response time <200 milliseconds; path deviation error: lane change scenario <0.5 meters, route change scenario <1.5 meters. The following test results were obtained: (1) Recognition performance: Test set (containing 2000 samples, covering rain, fog, night, and tunnel scenes): accuracy 98.2%, false detection rate <1.5%. Comparative experiment: single image recognition accuracy 89.3%, single sound recognition accuracy 72.6%.

[0056] (2) Avoidance efficiency: Urban road test (5 avoidance scenarios): average response time 185ms, avoidance success rate 100%. Highway test (speed 100km / h): lane change action completion time 2.8s, path deviation <0.5m.

[0057] (3) Resource consumption: Algorithm running platform: NVIDIA Orin (computing power 200TOPS), CPU utilization ≤35% The ambulance avoidance method proposed in this application involves acquiring image and sound data of the current vehicle. The image data is input into a preset ambulance image recognition model to obtain a first confidence level, and the sound data is input into a preset ambulance sound recognition model to obtain a second confidence level. The first and second confidence levels are fused to obtain a fused confidence level. It is then determined whether the fused confidence level is greater than a preset fusion threshold. If the fused confidence level is greater than the preset fusion threshold, the current vehicle speed and current road condition information are acquired. Based on the current vehicle speed and current road condition information, it is determined whether the current vehicle meets preset lane-changing avoidance conditions. If the current vehicle meets the preset lane-changing avoidance conditions, the current road condition information is input into a preset path planning model to obtain a target planned path, and the current vehicle is controlled to avoid the vehicle using the target planned path. This solves the problems of low reliability in ambulance recognition and inflexible avoidance strategies caused by single-modal perception. This application integrates multi-modal perception, possesses dynamic path planning capabilities, and has high response efficiency, which can improve the safe yielding capability in complex traffic environments.

[0058] Next, referring to the accompanying drawings, an ambulance avoidance device according to an embodiment of this application is described.

[0059] Figure 7 This is a block diagram of an ambulance avoidance device according to an embodiment of this application.

[0060] like Figure 7 As shown, the ambulance's obstacle avoidance device 10 includes: an acquisition module 100, an input module 200, and an obstacle avoidance module 300.

[0061] The acquisition module 100 is used to acquire the image data and sound data of the current vehicle.

[0062] The input module 200 is used to input image data into a preset ambulance image recognition model to obtain a first confidence level, and input sound data into a preset ambulance sound recognition model to obtain a second confidence level. The first confidence level and the second confidence level are fused to obtain a fused confidence level, and it is determined whether the fused confidence level is greater than a preset fusion threshold.

[0063] The obstacle avoidance module 300 is used to obtain the current vehicle speed and current road condition information when the fusion confidence is greater than the preset fusion threshold, determine whether the current vehicle meets the preset lane change and obstacle avoidance conditions based on the current vehicle speed and current road condition information, and input the current road condition information into the preset path planning model to obtain the target planned path when the current vehicle meets the preset lane change and obstacle avoidance conditions, and control the current vehicle to avoid obstacles by following the target planned path.

[0064] Optionally, in some embodiments, the avoidance module 300 includes: a judgment unit and a determination unit.

[0065] The judgment unit is used to calculate the minimum lane change distance based on the current vehicle speed, determine whether the lane line between the current vehicle and the adjacent lane is a dashed line based on the current road condition information, and determine whether the distance between the current vehicle and the target obstacle is greater than the minimum safe distance.

[0066] The determination unit is used to determine that the current vehicle meets the preset lane change and avoidance conditions when the lane change distance is greater than the minimum lane change distance, the lane line between the current vehicle and the adjacent lane is a dashed line, and the distance between the current vehicle and the target obstacle is greater than the minimum safe distance.

[0067] Optionally, in some embodiments, the input module 200 further includes a data re-acquisition unit and a fusion unit.

[0068] The data reacquisition unit is used to enable millimeter-wave radar-assisted verification when the fusion confidence is less than or equal to a preset fusion threshold. If the millimeter-wave radar-assisted verification passes, the acquisition frame rate of the image data is increased, and / or the audio analysis window of the sound data is extended to obtain new image data and new sound data.

[0069] The fusion unit obtains a new first confidence level and a new second confidence level based on the new image data and the new sound data, and obtains a new fusion confidence level based on the new first confidence level and the new second confidence level, and determines whether the new fusion confidence level is greater than a preset fusion threshold.

[0070] Optionally, in some embodiments, before inputting the current road condition information into a preset path planning model to obtain the target planned path, the avoidance module 300 includes: an acquisition unit, a division unit, a construction unit, an input unit, and a generation unit.

[0071] The acquisition unit is used to acquire historical path data and construct a dataset based on the historical path data.

[0072] The partitioning unit is used to divide the dataset into training, validation, and test sets based on a preset partitioning ratio.

[0073] The building unit is used to construct the target neural network. It is obtained by inputting the training set into the target neural network for training and obtaining the initial model parameters.

[0074] The input unit is used to input the validation set into the target neural network for performance evaluation based on the initial model parameters, and adjust the initial model parameters according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters.

[0075] The generation unit is used to input the test set into the target neural network based on the optimal model parameters to test the model, and obtain the preset path planning model when the test results meet the preset requirements.

[0076] It should be noted that the explanation of the above-described method for avoiding ambulances also applies to the ambulance avoidance device of this embodiment, and will not be repeated here.

[0077] The ambulance avoidance device proposed in this application acquires image and sound data of the current vehicle. The image data is input into a preset ambulance image recognition model to obtain a first confidence level, and the sound data is input into a preset ambulance sound recognition model to obtain a second confidence level. The first and second confidence levels are fused to obtain a fused confidence level. It is then determined whether the fused confidence level is greater than a preset fusion threshold. If the fused confidence level is greater than the preset fusion threshold, the current vehicle speed and current road condition information are acquired. Based on the current vehicle speed and current road condition information, it is determined whether the current vehicle meets preset lane-changing avoidance conditions. If the current vehicle meets the preset lane-changing avoidance conditions, the current road condition information is input into a preset path planning model to obtain a target planned path, and the current vehicle is controlled to avoid the vehicle using the target planned path. This solves the problems of low reliability in ambulance recognition and inflexible avoidance strategies caused by single-modal perception. This application integrates multi-modal perception, possesses dynamic path planning capabilities, and has high response efficiency, which can improve the safe yielding capability in complex traffic environments.

[0078] Figure 8 A schematic diagram of the structure of a vehicle provided in an embodiment of this application. The vehicle may include: The memory 801, the processor 802, and the computer program stored on the memory 801 and capable of running on the processor 802.

[0079] When the processor 802 executes the program, it implements the ambulance avoidance method provided in the above embodiments.

[0080] Furthermore, the vehicle also includes: Communication interface 803 is used for communication between memory 801 and processor 802.

[0081] The memory 801 is used to store computer programs that can run on the processor 802.

[0082] The memory 801 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0083] If the memory 801, processor 802, and communication interface 803 are implemented independently, then the communication interface 803, memory 801, and processor 802 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 8 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0084] Optionally, in a specific implementation, if the memory 801, processor 802, and communication interface 803 are integrated on a single chip, then the memory 801, processor 802, and communication interface 803 can communicate with each other through an internal interface.

[0085] The processor 802 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0086] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described ambulance avoidance method.

[0087] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0088] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0089] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0090] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0091] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it includes one or a combination of the steps of the method embodiments.

[0092] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for avoiding an ambulance, characterized in that, Includes the following steps: Acquire the current vehicle's image and audio data; The image data is input into a preset ambulance image recognition model to obtain a first confidence level, and the sound data is input into a preset ambulance sound recognition model to obtain a second confidence level. The first confidence level and the second confidence level are fused to obtain a fused confidence level, and it is determined whether the fused confidence level is greater than a preset fusion threshold. If the fusion confidence is greater than the preset fusion threshold, the current vehicle speed and current road condition information are obtained. Based on the current vehicle speed and current road condition information, it is determined whether the current vehicle meets the preset lane change and avoidance conditions. When the current vehicle meets the preset lane change and avoidance conditions, the current road condition information is input into the preset path planning model to obtain the target planning path, and the current vehicle is controlled to avoid the obstacle using the target planning path.

2. The method according to claim 1, characterized in that, The step of determining whether the current vehicle meets the preset lane change and avoidance conditions based on the current vehicle speed and the current road condition information includes: The minimum lane change distance is calculated based on the current vehicle speed. Based on the current road condition information, it is determined whether the lane line between the current vehicle and the adjacent lane is a dashed line, and it is determined whether the distance between the current vehicle and the target obstacle is greater than the minimum safe distance. When the lane change distance is greater than the minimum lane change distance, the lane line between the current vehicle and the adjacent lane is the dashed line, and the distance between the current vehicle and the target obstacle is greater than the minimum safe distance, the current vehicle is determined to meet the preset lane change and avoidance conditions.

3. The method according to claim 1, characterized in that, The step of determining whether the fusion confidence level is greater than a preset fusion threshold further includes: If the fusion confidence is less than or equal to the preset fusion threshold, millimeter-wave radar-assisted verification is enabled. If the millimeter-wave radar-assisted verification passes, the acquisition frame rate of the image data is increased, and / or the audio analysis window of the sound data is extended to obtain new image data and new sound data. A new first confidence level and a new second confidence level are obtained based on the new image data and the new sound data, and a new fusion confidence level is obtained based on the new first confidence level and the new second confidence level, and it is determined whether the new fusion confidence level is greater than the preset fusion threshold.

4. The method according to claim 1, characterized in that, Before inputting the current traffic information into the preset route planning model to obtain the target planned route, the process includes: Obtain historical path data and construct a dataset based on the historical path data; Based on a preset partitioning ratio, the dataset is divided into a training set, a validation set, and a test set. Construct a target neural network by inputting the training set into the target neural network for training to obtain initial model parameters; Based on the initial model parameters, the validation set is input into the target neural network for performance evaluation, and the initial model parameters are adjusted according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters. Based on the optimal model parameters, the test set is input into the target neural network for model testing, and when the test results meet the preset requirements, the preset path planning model is obtained.

5. A collision avoidance device for an ambulance, characterized in that, include: The acquisition module is used to acquire the image and sound data of the current vehicle; The input module is used to input the image data into a preset ambulance image recognition model to obtain a first confidence level, and input the sound data into a preset ambulance sound recognition model to obtain a second confidence level, fuse the first confidence level and the second confidence level to obtain a fused confidence level, and determine whether the fused confidence level is greater than a preset fusion threshold. The obstacle avoidance module is used to obtain current vehicle speed and current road condition information when the fusion confidence is greater than the preset fusion threshold, determine whether the current vehicle meets the preset lane change and obstacle avoidance conditions based on the current vehicle speed and the current road condition information, and when the current vehicle meets the preset lane change and obstacle avoidance conditions, input the current road condition information into the preset path planning model to obtain the target planned path, and control the current vehicle to avoid obstacles using the target planned path.

6. The apparatus according to claim 5, characterized in that, The avoidance module includes: The judgment unit is used to calculate the minimum lane change distance based on the current vehicle speed, determine whether the lane line between the current vehicle and the adjacent lane is a dashed line based on the current road condition information, and determine whether the distance between the current vehicle and the target obstacle is greater than the minimum safe distance. The determination unit is used to determine that the current vehicle meets the preset lane change and avoidance conditions when the lane change distance is greater than the minimum lane change distance, the lane line between the current vehicle and the adjacent lane is the dashed line, and the distance between the current vehicle and the target obstacle is greater than the minimum safe distance.

7. The apparatus according to claim 5, characterized in that, The input module further includes: The data reacquisition unit is used to enable millimeter-wave radar-assisted verification when the fusion confidence is less than or equal to the preset fusion threshold. If the millimeter-wave radar-assisted verification passes, the acquisition frame rate of the image data is increased and / or the audio analysis window of the sound data is extended to obtain new image data and new sound data. The fusion unit obtains a new first confidence level and a new second confidence level based on the new image data and the new sound data, and obtains a new fusion confidence level based on the new first confidence level and the new second confidence level, and determines whether the new fusion confidence level is greater than the preset fusion threshold.

8. The apparatus according to claim 5, characterized in that, Before inputting the current road condition information into the preset path planning model to obtain the target planned path, the avoidance module includes: An acquisition unit is used to acquire historical path data and construct a dataset based on the historical path data; A partitioning unit is used to divide the dataset into a training set, a validation set, and a test set based on a preset partitioning ratio. A construction unit is used to construct a target neural network, and to obtain initial model parameters by inputting the training set into the target neural network for training. The input unit is used to input the validation set into the target neural network for performance evaluation based on the initial model parameters, and adjust the initial model parameters according to the performance evaluation results until the joint loss function of the validation set converges to obtain the optimal model parameters. The generation unit is used to input the test set into the target neural network based on the optimal model parameters to test the model, and when the test results meet the preset requirements, obtain the preset path planning model.

9. A vehicle, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the ambulance avoidance method as described in any one of claims 1-4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the ambulance avoidance method as described in any one of claims 1-4.

Images